Stop Chatting. Start Briefing.

00:00Stop treating AI like a messy WhatsApp chat

I can almost guarantee that you are speaking to ChatGPT, Claude, Gemini, or whichever model that you use incorrectly. And the reason why I can say that because I bet you're one of those people who opens up, one of your AI chats and then you drop in a half formed thought like a, with a few typos and says, can you make this better? And expects the model to understand what's going on in your head. And if that's you, I'm talking to you and you're using one of the most capable models that's ever been put in our hands, like It's a messy WhatsApp chat.

A sentence here, a correction there, a new instruction. Five messages later, no clear outcome, and no shared definition of good.

00:54Why vague AI prompts fail

By the end of this session, you'll be able to take a vague request and turn it into a brief the tool can actually use and check that result against a clear standard in around 10 minutes or so. But here's what kind of makes that hard. Is that bad? AI doesn't always look bad. And it could. And the reason for this is it can be grammatically perfect. It can have a strong subject line, tidy paragraphs, a clear next step. It can sound confident, it can look complete. And you know, kind of people see all of that and they just assume that the tool understood them and got what they were trying to do. But the reality is, is that smooth is not the same as useful, and complete isn't the same as right for the job.

02:12Smooth AI output vs useful work

And looking right just doesn't mean that it always is right.

03:06The generic ChatGPT prompt test

Imagine. I ask AI, write me an email to promote my new masterclass about getting better results from AI. It's for women in business. Make it warm, punchy, but not too salesy.

The result is decent. It tells me that AI can save time, sharpen my thinking, and help me do better work. It promises confidence, clarity, and purpose. Practical guidance. There's no jargon. And it ends with two possible calls to action. There's nothing obviously wrong with it, but it could be selling almost any AI class. women in business is still the audience. the model has invented the promise. It's guessed the outcome. It's handed me two calls to action because I never made those calls myself. no name, no title, no black dress idea. No reason this masterclass is any different from the rest. And the output, it's not bad. It's plausible. And looking right doesn't make it right.

04:17The invisible decisions AI makes

Before I send this email, I need to ask a different question because what did the model have to decide that I didn't? It had to decide which women I meant because a, senior executive or brand freelancer, a brand new freelancer or a, female founder, all already using AI, they don't all need the same message. It had to decide whether, the email was meant to build brand awareness registrations, whether it was meant to sell membership, or whether I just wanted it to get a reply. It had to invent the differentiator. guess what warm sounds like in my voice. choose whether, you know, which claims were okay to make. It had to decide whether its first draft was even finished.

And that's why mediocre AI is so easy to accept, because the missing decisions are basically invisible. You see the polished answer and you don't see the, like the trail of guesses that were taking place underneath it. So when you chat to AI like it's a long running chat group, like a long running WhatsApp chat, you know the backstory and the tool doesn't, and you're answering the job in your head and the model's answering the words that you actually gave it, so the output can be correct and still be wrong for the job.

05:54Princess Diana’s black dress as a briefing example

The easiest way I know how to explain the difference between asking and briefing is with a princess, a little black dress and an imaginary stylist. Because Princess Diana understood that clothes could talk before she ever said a word. Whatever you think of the phrase revenge dress, the images stuck because the dress wasn't just an item of clothing, it did a job. On 29 June 1994, Diana attended the Serpentine Gallery event in London and That same evening, a television documentary aired in which her separated husband, the then Prince Charles, admitted to an affair. And Diana arrived in a short, off the shoulder, Christina Stambolian, dress, little black dress. And the photographs became part of the story of that night.

07:43Why “a black dress” names an object, not a job

Now, I want to be very clear that what follows is completely hypothetical. it's a teaching exercise, right? So I'm not reconstructing the real conversation between Diana and a stylist. So I, want you to just imagine that before that evening that Diana had said this. I need a black dress. A good stylist could come back with a ball gown, a, cocktail dress, something safe or even something daring. And every single one of those would answer that request. That's the point. A black dress describes an object, but it says almost nothing about the job. What's the occasion? What do the photographs need to say? What would be too safe? What would tip, over into costume?

What does the dress need to let her do? Those questions aren't extra detail. They are the job. So imagine now that the stylist gets a proper brief and the brief gives her context so she understands that it's, a high profile evening event under intense scrutiny from the global world's press. The result she wants is a woman who looks composed, who looks independent and in control. And then the boundaries that she's working within are the dress needs to be black, it needs to be elegant, it needs to look bold in photography photographs, and is comfort comfortable enough to move around in, to greet people in, and but never, ever looks theatrical. And then a reference photo could carry across that the, you know, the dress needs to have a clean line or, and shows like a confident neckline and a lack of fuss in photos.

09:05Five questions every AI brief should answer

And then the fail conditions that the stylist could be working with could rule out anything that reads as mourning or apology or costume or discomfort or anything that's like so cautious that it causes her to kind of disappear and not present as that bold woman, at, the event. Now, the stylist now is not choosing from every single black dress in the world. The stylist's brief has stripped out thousands of answers that are highly plausible but would be wrong for this particular job. Useful brief answers five kinds of questions. What is the context? What results must this create? What are the boundaries? What reference would clarify the direction? And what would make the choice completely wrong?

I want you to notice what's happened. The stylist hasn't been told every stitch to sew or every move to make, she's been given enough to use her own judgment and to choose well. And so what that means is that that's exactly what a good brief does. It allows you to make right choices based on having a full scope of work. It doesn't take away, from the expertise of the person or the tool doing the work. It makes the job clear so that the expertise can be pointed at the right problem. And then giving an example doesn't mean copy this exactly. It doesn't mean that. What it actually means is carry this quality across. And so if there is no useful reference, my advice is don't invent one.

You know, don't invent one just to fill the framework. The only question is whether an example will make the work actually better. So to remind us, the request was a little black dress. The job was to build a public image that showed confidence under the world's press scrutiny. And the black dress describes the object. it doesn't explain the job thing. Happens every day with AI. You write a LinkedIn post, create a proposal, make this page better, help me with my strategy. And those are phrases that name an object or an activity. They don't hand over a job. So when the job is missing, AI doesn't usually put up its hand and go, hey, Michelle, you haven't made enough decisions here.

10:01AI literacy is judgement, not just prompting

It just fills in the gaps. It picks an audience, an aim, a tone, a structure, a standard even. And it does it all for you. And it's not necessarily what you want it to do. So request names a thing, whereas a brief hands over a job. So before you accept anything AI gives, you ask this. What did it have to guess? Is it right for the result I actually need? And what tells me it's any good? It's a very different test from, does this sound good, sounding good? Exactly how, like plausible work actually slips past you. If you can't name the result, you can't judge whether it reaches its destination. And if you can't name the standard, you're picking between drafts on a feeling.

Literally, you're picking on feelings. And if you can't see what the model guess, then you can't see where you. Where your own personal judgment is missing from the equation. So basically, AI literacy isn't. It isn't knowing how to get an answer, it's knowing how to tell whether the answer's any good. And the first skill isn't actually asking, it's noticing. So once you see the hidden decisions, you can't stop leaving them to the model. And that's exactly what the BRIEF method helps you do. BRIEF stands for background, result, instructions, example and fail conditions. It's deliberately simple because you can use it with ChatGPT, you can use it with Claude, you can use it with Gemini, another AI tool, or you can use it with a human being.

11:20BRIEF: Background

Because it's not really a prompting framework. it's a method for thinking through a job before you hand it over. So brief is the thinking, the prompt is simply the handoff. Background is the context. The model cannot safely guess. So who is this for? What are we working on? What has already happened? What has been decided? What source material matters? What constraints come from the real situation? This is not an invitation to paste your autobiography into every chat. It's a, relevant backstory for this job. Right, so with the dress, the background was the event, that the, the scrutiny and that the public context that was taking place around that evening. And with an email, it could be, it might be the relationship with the reader, you know, the recipient, what, what they already know, why the message is being sent now and what promises has already been made.

And if you skip background, AI doesn't kind of just leave you with, a, neat little blank space where your expertise should be, it will substitute it with assumptions. So background is context, not an autobiography. Result is what the finished work must achieve. So this is where a lot of requests basically stay too shallow. Because write an email is a task. Write an email that helps existing clients understand a price increase, that helps them keep their confidence and know exactly what to do. If they got a question. Now, that is a result. So name a deliverable, yes, but also name the job it has to do. you want to know what response it should create, what, decision should it help with, and you want to know what should happen next.

12:22BRIEF: Result

Basically, the clearer the result, the easier it is to judge what you get back. Without it, you're just picking the version that sounds nicer, and that's not really a standard. So don't give AI an activity, give it a result. Instructions are the requirements and boundaries that shape how work should be done. Format, length, tone, what must be in it, what to leave out the channel it's going to appear in, the source material it has to use. Basically, good instructions, take away the choices that are, irrelevant or maybe even sometimes dangerous. Make it punchy, is not a boundary, whereas short use short paragraphs. one clear call to action in UK English, no inflated claims.

13:28BRIEF: Instructions

With no force urgency is so telling AI that accuracy is critical doesn't make an unsupported claim, you know, true. If the job needs a document or it needs data, live, research, or at all, then give that to the model or tell it to flag what it can't check. Good instructions take away the wrong choices. Example shows the model what good looks like when the words alone are not enough. You might share a piece of your own writing or an old proposal, a structure you want to follow or reference that shows the level of detail that you expect. But make it like this is weak briefing. You need to say what it is that you want it to borrow.

13:54BRIEF: Example

So it could be the calm tone, or is it the short opening, the way that evidence gets introduced, or the shape of the argument and name what it is that shouldn't be copied. Example is a decision inside the method. So if a useful example would make the work better, give it and say why it's useful as well. If there's no, genuinely useful example, leaving the field blank is actually, better than making one up because that just adds noise. So the main takeaway is the thinking isn't optional. The example itself has to earn its place. Use the example to carry across quality, not to outsource your taste. Fail conditions define what would make the output usable even if it looked polished.

15:05BRIEF: Fail Conditions

What must not happen, what must be checked, what's to be kept, what would make you reject the work. What should AI flag instead of inventing? for a proposal, a fail condition might be inventing client result. Or it could be making a promise that delivery can't keep. Or for a, customer email, it might be sounding apologetic or reopening a decision that's already been settled. Or leaving a customer with no clear next steps. For the dress, it was mourning, you know, it was anything that read as costume or discomfort or caution, when the job, you know, needed control. So AI is very, very good at producing something that looks finished. And you have to tell it what looks you, you have to tell it what wrong looks like before the polish kind of hides the failure and fail.

16:03Live comparison: vague prompt vs complete brief

What fail conditions do is they stop finished looking work from passing the wrong test. And that completes the BRIEF method, background, result instructions, example and fail conditions. Let's prove it with the same task from this masterclass. I want an email that invites experienced women founders and leaders to stop chatting and start briefing. I'm keeping the tool the same, the task the same, and the date the same. The only thing that is changing is the quality of the brief. So first I'll give ChatGPT the same WhatsApp style request that you saw earlier. Write me an email to promote my new masterclass about getting better results from AI. It is for women in business. Make it warm and punchy, but not too salesy.

Email is fluent. It says AI can save time, sharpen thinking and help create better work. It promises confidence, clarity and purpose. And it offers practical, practical guidance without any jargon. Selling almost any AI masterclass, as we identified before, the audience is still just women in business. The model invented the promise, it guessed the outcome. It's handed me two possible calls to action. It doesn't know who's sending the email and there's stuff missing, like there's no title, there's no black dress idea, there's no argument that it's actually its own. there's nothing obviously, broken about it. And that's exactly why it's a problem. That's why this matters. It's because the output is good enough to send.

17:13Building the brief step by step

It's like plausible enough to send, let's say. But it's so generic and you know, it feels like a waste of opportunity to me because it's too generic. So same tool, same task. you know, but I'm gonna change the brief. Okay, so watch what changes When I make the assignment visible, what I'm going to do for the interest of time and because I'm not a fast typist, I am going to copy and paste some what I've already typed out into the boxes. So for background, I identify myself, the experienced founders and leaders that I want the email to speak to. And you know, and I'm, you know, what they already use and basically what the masterclass is not.

And then we click next I'm just going to copy over result. so in result, I'm defining one email that is I want it to be, under 180 words that makes the right reader recognize her behavior and understand the promise and has one clear action for instructions. Let me just grab that. I've made the voice structure and the boundaries very explicit. So it needs to be in UK English, first person, etc. Etc. We press next. For example, I am deliberately leaving this blank. because of the nature of this type of email, it's going to be, you know, this hypothetical email is going out to a warm email list and the background and the voice instructions are sufficient for this particular task.

So I don't need to manufacture an example that wouldn't necessarily materially improve the result. But generally speaking it makes sense to give it an example to show what good, looks like. So I'm going to press next and then for F, which are the fail conditions, I am, giving it some detailed conditions so it knows what, what to avoid. You know, nothing. We don't want any generic. It could be any AI event. We don't want cliche words and cliche phrases. So I kind of get rid of anything that is like nah, we don't want that. And then we're going to run the brief check. And here's the result. Okay, it's almost ready. It is giving me a recommendation.

So I'm just going to quickly review the recommendation because it has identified one decision that I haven't made. So I'm just going to see if I agree with the, recommendation. Okay, I do. So I'm going to add extra criteria and run the brief check again. Okay, great. So that has passed now and it's sufficient to add it, to my AI. In this case, I'm going to be using ChatGPT, so I'm going to copy it over, just quickly and expect, the full brief. I now have a full scope of work to delegate over to AI and I am going to copy the brief. Now I can hand the assembled brief over to ChatGPT.

19:31Comparing both ChatGPT outputs

The response is no longer guessing what I mean. It still needs my judgment, but it understands the assignment now. Same tool, same task, different brief. The difference is is that ChatGPT didn't suddenly become more intelligent. The difference is, is that the assignment became more legible. Now, compare both outputs against the same quality bar. The second email speaks to a specific audience. It names the messy chat behavior, it makes the masterclass distinctive through the black dress idea, and it has one clear call to action and it stays inside the claims and tone that I approved. Keep what's already working, what must not be lost, then change what specifically needs to be different then, because why does that change matter to the result?

20:23Keep, Change, Because, Next Version Must

And then finish with next version must. What does the revision have to get right before you accept it? So keep change because next version must that turns your feedback into direction instead of just another trail of corrections. Here's what I want you to do next. Take 10 minutes and re brief one real piece of work. Find a recent AI output that looked decent but wasn't quite right. And then name the guesses the model made because you didn't give it the decisions. Then I, want you to brief the task properly using all five parts of the method. So that's background, result instructions, example and fail conditions. Then put that new output next to the original and then judge them against the same standard.

25:10Your 10-minute re-brief challenge

So don't ask which one's nicer, ask which one understood the job and then prove it. The two key takeaways. First, I want you to think before you delegate. Like just take that extra five to ten minutes up front to think through before you hand stuff over to AI. You know, what you're essentially aiming for with the first draft is to get it set, you know, to almost one shot. It's 70 to 80% of the way there because you did that upfront work. And then that leaves the, remaining 20% of for your judgment, for iteration rather than trying to rescue a messy chat approach. Right? The second thing I'd love you to take away is to literally download it and put it to work because, you know, I don't want you just kind of nodding along and agreeing with today and then just never using it.

26:45See what AI had to guess

So head on over to aipowerwomen.com/brief and download the free brief resource that's there. Use it and rebrief one piece of real work this week. In fact, do it today. And if you want help building a brief, so you can spot where it's too thin, where it's where you know you want it to get to a position where the handoff is really clear. For ChatGPT, Claude and Gemini, and any other AI tools that you use, then the guided brief builder is inside the AI Power Women library alongside other practical tools, that are from female founders, from other female founders that are built to get real business work done.

And so that's at, aipowerwomen.com/brief, ChatGPT, Claude, Gemini. They, they don't need you to sound technical. What they need you to do is to make the decisions that they can't. I want you to really take that in. I want you to think of it this way. You'd brief a stylist for the most important night of your life and you wouldn't hand over a messy WhatsApp group chat and just hope, right? What you do is you tell her the occasion, the result that you want, what you count as wrong, you know, what would, what it would look like if it was wrong. And then you let her know, what good looks like as well, maybe through an example.

And then you just let her do what she's brilliant at and let the whole thing her just get on with it and come up with a beautiful piece for you. And that should be the shift. You know, stop chatting, start briefing.